A Lightweight Robust Deep Learning Model Gained High Accuracy in Classifying a Wide Range of Diabetic Retinopathy Images

IEEE Access, Год журнала: 2023, Номер 11, С. 42361 - 42388

Опубликована: Янв. 1, 2023

Diabetic retinopathy (DR) is a common complication of diabetes mellitus, and retinal blood vessel damage can lead to vision loss blindness if not recognized at an early stage. Manual DR detection using large fundus image data time-consuming error-prone. An effective automatic system be significantly faster potentially more accurate. This study aims classify images into five classes, deep learning methods, with the highest possible accuracy lowest computational time. Three distinct datasets, APTOS, Messidor2, IDRiD, are merged, resulting in 5,819 raw images. Before training model, various preprocessing techniques applied remove artifacts noise from improve their quality. augmentation techniques: geometric, photometric, elastic deformation, used create balanced dataset. A shallow convolutional neural network (CNN) developed three blocks layers maxpool categorical cross-entropy function, Adam optimizer, 0.0001 rate, 64 batch size as base this also employed determine best method for further processing. optimize performance then conducted by changing different components hyperparameters our proposed RetNet-10 model. Six cutting-edge models comparison. Our model performed best, testing 98.65%. MobileNetV2, VGG16, Xception, VGG19, InceptionV3 ResNet50 achieved accuracies 91.42%, 90.16%,89.57%, 88.21%, 87.68% 87.23%, respectively. The trained several k values assess its robustness. After processing augmentation, combined dataset, fine-tuning outperformed other automated methods diagnosis.

Язык: Английский

---

Redefining Retinal Lesion Segmentation: A Quantum Leap With DL-UNet Enhanced Auto Encoder-Decoder for Fundus Image Analysis

IEEE Access, Год журнала: 2023, Номер 11, С. 70853 - 70864

Опубликована: Янв. 1, 2023

The first diagnosis of diabetic retinopathy (DR) must include lesion segmentation. As it takes a lot time and effort to label lesions, automatic segmentation methods have be created manually. degree the retina's degenerative lesions determines how severe is. A major influence is on early detection illness treatment DR. To reliably identify sites related various abnormalities in retinal fundus pictures, deep learning algorithms are crucial. Additionally, utilizing patch-based analysis, convolutional neural network constructed. In this study, encoder-decoder networks along with channel-wise spatial Attention Mechanisms proposed. IDRiD dataset, which includes hard exudate segmentations, used train evaluate architecture. method, image patches using sliding window technique. determine effectiveness recommended strategy, thorough experiment was conducted IDRiD. order predict sorts trained analyses picture creates probability map. This technique's efficacy supremacy confirmed by expected accuracy 99.94 %. findings show significantly enhanced performance terms when compared prior research comparable tasks.

Язык: Английский

---

Classification and Segmentation of Diabetic Retinopathy: A Systemic Review

Applied Sciences, Год журнала: 2023, Номер 13(5), С. 3108 - 3108

Опубликована: Фев. 28, 2023

Diabetic retinopathy (DR) is a major reason of blindness around the world. The ophthalmologist manually analyzes morphological alterations in veins retina, and lesions fundus images that time-taking, costly, challenging procedure. It can be made easier with assistance computer aided diagnostic system (CADs) are utilized for diagnosis DR lesions. Artificial intelligence (AI) based machine/deep learning methods performs vital role to increase performance detection process, especially context analyzing medical images. In this paper, several current approaches preprocessing, segmentation, feature extraction/selection, classification discussed This survey paper also includes detailed description datasets accessible by researcher identification existing limitations challenges addressed, which will assist invoice researchers start their work domain.

Язык: Английский

---

Swin Transformer-Based Segmentation and Multi-Scale Feature Pyramid Fusion Module for Alzheimer’s Disease with Machine Learning

International Journal of Online and Biomedical Engineering (iJOE), Год журнала: 2023, Номер 19(04), С. 22 - 50

Опубликована: Апрель 3, 2023

Alzheimer Disease (AD) is the ordinary type of dementia which does not have any proper and efficient medication. Accurate classification detection AD helps to diagnose in an earlier stage, for that purpose machine learning deep techniques are used observers both normal abnormal brain accurately detect early. For accurate AD, we proposed a novel approach detecting using MRI images. The work includes three processes such as tri-level pre-processing, swin transfer based segmentation, multi-scale feature pyramid fusion module-based detection.In noises removed from images Hybrid Kuan Filter Improved Frost (HKIF) algorithm, skull stripping performed by Geodesic Active Contour (GAC) algorithm removes non-brain tissues increases accuracy. Here, bias field correction Expectation-Maximization (EM) intensity non-uniformity. After completed initiate segmentation process Swin Transformer Segmentation Modified U-Net Generative Adversarial Network (ST-MUNet) segments gray matter, white cerebrospinal fluid considering cortical thickness, color, texture, boundary information that, extraction Multi-Scale Feature Pyramid Fusion Module VGG16 (MSFP-VGG16) extract features accuracy, on extracted image classified into classes (AD), Mild Cognitive Impairment, Normal. simulation this research conducted Matlab R2020a tool, performance evaluated ADNI dataset terms specificity, sensitivity, confusion matrix, positive predictive value.

Язык: Английский

---

Internet of Things and Deep Learning Enabled Diabetic Retinopathy Diagnosis Using Retinal Fundus Images

IEEE Access, Год журнала: 2023, Номер 11, С. 27590 - 27601

Опубликована: Янв. 1, 2023

Recently, the Internet of Things (IoT) and computer vision technologies find useful in different applications, especially healthcare. IoT driven healthcare solutions provide intelligent for enabling substantial reduction expenses improvisation service quality. At same time, Diabetic Retinopathy (DR) can be described as permanent blindness eyesight damage because diabetic condition humans. Accurate early detection DR could decrease loss damage. Computer-Aided Diagnoses (CAD) model based on retinal fundus image is a powerful tool to help experts diagnose DR. Some traditional Machine Learning (ML) diagnoses has currently existed this study. The recent developments Deep (DL) its considerable achievement over conventional ML algorithms applications make it easier design effectual diagnosis model. With motivation, paper presents novel DL enabled retinopathy (IoTDL-DRD) using images. presented – Diagnosis technique utilizes devices data collection purposes then transfers them cloud server process them. Followed by, images are preprocessed remove noise improve contrast level. Next, mayfly optimization region growing (MFORG) segmentation utilized detect lesion regions image. Moreover, densely connected network (DenseNet) feature extractor Long Short Term Memory (LSTM) classifier used effective diagnosis. Furthermore, parameter LSTM method carried out by Honey Bee Optimization (HBO) algorithm. For evaluating improved diagnostic outcomes IoTDL-DRD technique, comprehensive set simulations were out. A wide ranging comparison study reported superior performance proposed method.

Язык: Английский

---

Effect of Changing Targeted Layers of the Deep Dream Technique Using VGG-16 Model

International Journal of Online and Biomedical Engineering (iJOE), Год журнала: 2023, Номер 19(03), С. 34 - 47

Опубликована: Март 14, 2023

The deep dream is one of the most recent techniques in learning. It used many applications, such as decorating and modifying images with motifs simulating patients' hallucinations. This study presents a model that generates using convolutional neural network (CNN). Firstly, we survey layers each block network, then choose required layers, extract their features to maximize it. process repeats several iterations needed, computes total loss, extracts final images. We apply this operation on different two times; former low-level latter high-level layers. results applying are different, where resulting image from clearer than those Also, loss ranges between 31.1435 31.1435, while upper 20.0704 32.1625.

Язык: Английский

---

A Lightweight Diabetic Retinopathy Detection Model Using a Deep-Learning Technique

Diagnostics, Год журнала: 2023, Номер 13(19), С. 3120 - 3120

Опубликована: Окт. 3, 2023

Diabetic retinopathy (DR) is a severe complication of diabetes. It affects large portion the population Kingdom Saudi Arabia. Existing systems assist clinicians in treating DR patients. However, these entail significantly high computational costs. In addition, dataset imbalances may lead existing detection to produce false positive outcomes. Therefore, author intended develop lightweight deep-learning (DL)-based DR-severity grading system that could be used with limited resources. The proposed model followed an image pre-processing approach overcome noise and artifacts found fundus images. A feature extraction process using You Only Look Once (Yolo) V7 technique was suggested. provide sets. employed tailored quantum marine predator algorithm (QMPA) for selecting appropriate features. hyperparameter-optimized MobileNet V3 utilized predicting severity levels generalized APTOS EyePacs datasets. contained 5590 images, whereas included 35,100 outcome comparative analysis revealed achieved accuracy 98.0 98.4 F1 Score 93.7 93.1 datasets, respectively. terms complexity, required fewer parameters, floating-point operations (FLOPs), lower learning rate, less training time learn key patterns nature can allow healthcare centers serve patients remote locations. implemented as mobile application support future, will focus on improving model's efficiency detect from low-quality

Язык: Английский

---

Deep learning innovations in diagnosing diabetic retinopathy: The potential of transfer learning and the DiaCNN model

Computers in Biology and Medicine, Год журнала: 2023, Номер 169, С. 107834 - 107834

Опубликована: Дек. 11, 2023

Язык: Английский

---

Classification of Diabetic Retinopathy by Deep Learning

International Journal of Online and Biomedical Engineering (iJOE), Год журнала: 2024, Номер 20(01), С. 74 - 88

Опубликована: Янв. 12, 2024

Diabetic retinopathy (DR), which is a leading cause of adult blindness, primarily affects individuals with diabetes. The manual diagnosis DR, the assistance an ophthalmologist, has proven to be time-consuming and challenging process. Late detection DR significant factor contributing progression disease. To address this issue, present study utilizes deep learning (DL) transfer algorithms analyze different stages precisely detect condition. Using large dataset comprising approximately 60,000 images, employs ResNet-101, DenseNet121, InceptionResNetV2, EfficientNetB0 DL models automatically assess DR. Images patients’ eyes are inputted into models, architectures adapted extract relevant features from eye images. study’s findings demonstrate that DenseNet121 outperforms in accurately classifying five accuracy was 97%, 96%, 95%, 94%, respectively. These results underscore effectiveness achieving accurate comprehensive classification retinitis pigmentosa. By enabling timely application techniques significantly contributes field ophthalmology, facilitating improved treatment decisions for patients.

Язык: Английский

---

An Integrated Multimodal Deep Learning Framework for Accurate Skin Disease Classification

International Journal of Online and Biomedical Engineering (iJOE), Год журнала: 2024, Номер 20(02), С. 78 - 94

Опубликована: Фев. 14, 2024

In order to effectively treat skin diseases, an accurate and prompt diagnosis is required. this article, a novel method for classifying disorders using multimodal classifier presented. The proposed utilizes multiple information sources enhance the accuracy of disease classification. It incorporates images lesions patient-specific data. simultaneously classifies diseases by combining image structured data inputs. effectiveness was evaluated ISIC 2018 dataset, which includes clinical seven categories diseases. results indicate that model outperforms conventional single-modal single-task classifiers, achieving 98.66% classification 94.40% addition, we compare performance with other methodologies, demonstrating its superiority. Despite yielding promising results, has limitations in terms requirements generalizability. Future research directions include incorporating additional sources, investigating genetic integration, applying various medical conditions. This study illustrates potential integrating techniques transfer learning deep neural networks cutaneous

Язык: Английский

---